Job YAML

Minimalist Job.yaml

job.yaml

# Local directory where your source code resides.
# It should be the relative path to this job yaml file.
# If your job doesn't contain any source code, it can be empty.
workspace: hello_world

# Bootstrap shell commands which will be executed before running entry commands.
# Support multiple lines, which can be empty.
bootstrap: |
  pip install -r requirements.txt
  echo "Bootstrap finished."

# Running entry commands which will be executed as the job entry point.
# If an error occurs, you should exit with a non-zero code, e.g. exit 1.
# Otherwise, you should exit with a zero code, e.g. exit 0.
# Support multiple lines, which can not be empty.
job: |
    echo "Hello, Here is the launch platform."
    echo "Current directory is as follows."
    pwd
    python hello_world.py
  
computing:
  minimum_num_gpus: 1      # minimum # of GPUs to provision
  # max cost per hour of all machines for your job.
  # E.g., if your job are assigned 2 x A100 nodes (8 GPUs), each GPU cost $1/GPU/Hour, "maximum_cost_per_hour" = 16 * $1 = $16
  maximum_cost_per_hour: $1.75

Tip

For most cases, you just need to use the above minimalist job.yaml with following four properties:

1. workspace, It is the local directory where your source code resides.
2. job, It is the running entry command which will be executed as the job entry point.
3. bootstrap, It is the bootstrap shell command which will be executed before running entry commands.
4. computing, It is the computing resource configuration for your job.

Fully loaded Job.yaml

Below is job.yaml loaded with all properties. You can use it as a reference to create your own job.yaml tailored to your specific needs.

job.yaml

fedml_env:
  project_name: 

# Local directory where your source code resides.
# It should be the relative path to this job yaml file or the absolute path.
# If your job doesn't contain any source code, it can be empty.
workspace: hello_world

# Running entry commands which will be executed as the job entry point.
# If an error occurs, you should exit with a non-zero code, e.g. exit 1.
# Otherwise, you should exit with a zero code, e.g. exit 0.
# Support multiple lines, which can not be empty.
job: | 
    echo "Hello, Here is the launch platform."
    echo "Current directory is as follows."
    pwd
    python hello_world.py

# Bootstrap shell commands which will be executed before running entry commands.
# Support multiple lines, which can be empty.
bootstrap: |
  pip install -r requirements.txt
  echo "Bootstrap finished."

computing:
  minimum_num_gpus: 1             # minimum # of GPUs to provision

  # max cost per hour of all machines for your job. 
  # E.g., if your job are assigned 2 x A100 nodes (8 GPUs), each GPU cost $1/GPU/Hour, "maximum_cost_per_hour" = 16 * $1 = $16
  maximum_cost_per_hour: $1.75
  
  allow_cross_cloud_resources: false # true, false
  device_type: GPU              # options: GPU, CPU, hybrid
  resource_type: A100-80G       # e.g., A100-80G, please check the resource type list by "fedml show-resource-type" or visiting URL: https://tensoropera.ai/accelerator_resource_type
  
job_type: train              # options: train, deploy, federate
framework_type: fedml        # options: fedml, deepspeed, pytorch, general

# train subtype: general_training, single_machine_training, cluster_distributed_training, cross_cloud_training
# federate subtype: cross_silo, simulation, web, smart_phone
# deploy subtype: none
job_subtype: general_training

# Running entry commands on the server side which will be executed as the job entry point.
# Support multiple lines, which can not be empty.
server_job: |
    echo "Hello, Here is the server job."
    echo "Current directory is as follows."
    pwd
    
# If you want to use the job created by the MLOps platform,
# just uncomment the following three, then set job_id and config_id to your desired job id and related config.
#job_args:
#  job_id: 2070
#  config_id: 111

# If you want to create the job with specific name, just uncomment the following line and set job_name to your desired job name.
#job_name: cv_job

# If you want to pass your API key to your job for calling TensorOpera APIs, you may uncomment the following line and set your API key here.
# You may use the environment variable FEDML_RUN_API_KEY to get your API key in your job commands or scripts.
#run_api_key: my_api_key

# If you want to use the model created by the MLOps platform or create your own model card with a specified name,
# just uncomment the following four lines, then set model_name to your desired model name or set your desired endpoint name
#serving_args:
#  model_name: "fedml-launch-sample-model" # Model card from MLOps platform or create your own model card with a specified name
#  model_version: "" # Model version from MLOps platform or set as empty string "" which will use the latest version.
#  endpoint_name: "fedml-launch-endpoint" # Set your end point name which will be deployed, it can be empty string "" which will be auto generated.

# Dataset related arguments
fedml_data_args:
  dataset_name: mnist
  dataset_path: ./dataset
  dataset_type: csv
  
# Model related arguments
fedml_model_args:
  input_dim: '784'
  model_cache_path: /Users/alexliang/fedml_models
  model_name: lr
  output_dim: '10'